Neural architectures for sensorfusion in speechrecognition